AI, EDR, and Hacking Things – PSW #896

Paul Asadoorian

1. Finding Critical Bugs in Adobe Experience Manager › Searchlight Cyber
2. A Story About Bypassing Air Canada’s In-flight Network Restrictions
3. IAmAntimalware: Inject Malicious Code Into Antivirus
4. Urgent: How Hackers Use eBPF to Evade Detection
5. IDA tips for reversing U-Boot
6. When weaponized AI can dismantle patches in 72 hours, kernel security needs to deliver
7. MEDICAL DEVICE Security Analysis
8. SonicWall Confirms Breach Exposing All Customer Firewall Configuration Backups
9. How to protect your car from hacking
10. Bash a newline: Exploiting SSH via ProxyCommand, again (CVE-2025-61984)
11. goichot/OverLAPS: Supporting PoCs and scripts for my talk “OverLAPS: Overriding LAPS Logic”
12. This ILLEGAL Device Instantly KILLS All Network & TV Signals
13. Taking remote control over industrial generators
14. TOTOLINK X6000R: Three New Vulnerabilities Uncovered
15. John Kristoff – Journeys in Hosting 1/x – Precomputed SSH Host Keys
16. Your Vulnerability Scanner Might Be Your Weakest Link

    This is an important point in the article: "For example, across multiple widely deployed EDR products, none of the credential dumping techniques we described earlier generated alerts during our experiments. Detection rates were effectively zero. This strongly suggests that many Linux EDR solutions are optimized to check compliance boxes (e.g., “Linux support available”) rather than to deliver feature parity with their Windows agents, which tend to be far more mature and capable of detecting common attacker tactics, techniques, and procedures (TTPs)." - We do lack great EDR on Linux, vendors please step up!

17. I’ve Written About Loads of Scams. This One Almost Got Me.

    Zelle is the Mos Eisley of banking; no one should ever use it again, unless you want to be part of a scam and send random people money. This scam is typical, somewhat smart, and likely works enough to warrant the effort.

18. ‘You’ll never need to work again’: Criminals offer reporter money to hack BBC

    This is my: "What if Russians drop off a bag of money scenario" that is actually happening in the real world. Again, it seems to work well enough that attackers are investing in it, which means users go to the dark side more than we like to think.

19. Microsoft, SentinelOne, and Palo Alto Networks Withdraw from 2026 MITRE ATT&CK Evaluations

    What in the world is going on here? What is wrong with the Mitre evaluations? The state of EDR is getting really interesting, with so many bypasses coming out all the time, where does that leave us?

20. Getting Started with the Raspberry Pi for Hacking: Using Spiderfoot for OSINT Data Gathering

    Uhm, I'm just here to point out that you do not need a RPI to run Spiderfoot, in fact, I would recommend against it. Run it on a real machine, its an awesome tool as I tested it out last week and plan to use it more often for OSINT. I highly recommend spending some time to get all the API keys from all the sources as well. Don't worry about running it on a PI, it runs just fine in a Python virtual environment. However, I did run into dependency issues and ended up having to run a newer version of one of the libraries but I have not noticed any resulting issues or errors.

21. EDR-Freeze: A Tool That Puts EDRs And Antivirus Into A Coma State

    EDR-Freeze is a tool and technique that leverages Windows internals,specifically the MiniDumpWriteDump function and WerFaultSecure.exe,to temporarily suspend the processes of EDR (Endpoint Detection and Response) and antivirus programs. This is achieved without using vulnerable third-party drivers (as in BYOVD, Bring Your Own Vulnerable Driver attacks), but by exploiting built-in Windows debugging mechanisms in user mode.

    • The MiniDumpWriteDump function, used to take memory snapshots for debugging, automatically suspends all threads of a target process during operation.
    • WerFaultSecure.exe, a built-in Windows tool, can be run at a privileged “PPL” (Protected Process Light) level, allowing access to processes normally protected from tampering—including antivirus and EDR.
    • By carefully synchronizing the creation and suspension of WerFaultSecure.exe (using CreateProcessAsPPL and the PROCESSSUSPENDRESUME privilege), attackers can put targeted security processes into a prolonged dormant state via a race condition.
22. One Token to rule them all – obtaining Global Admin in every Entra ID tenant via Actor tokens

    A critical vulnerability in Microsoft Entra ID (formerly Azure AD) allowed attackers to gain Global Admin access to any tenant using “Actor tokens,” which are undocumented tokens used for internal service-to-service authentication. These tokens bypassed all security controls—including Conditional Access—and exploited a flaw in the legacy Azure AD Graph API that failed to properly validate the tenant origin. By crafting a token with a victim tenant’s ID and the “netId” of any user, an attacker could impersonate any user, including Global Admins, and gain complete control without generating logs in the victim tenant.

23. GitHub – ZerkerEOD/krakenhashes

    I just found a use for a bunch of older computers: "KrakenHashes is a distributed password cracking system designed for security professionals and red teams. The platform coordinates GPU/CPU resources across multiple agents to perform high-speed hash cracking using tools like Hashcat through a secure web interface. Think of KrakenHashes as a full management system for hashes during, after and before (if a repeat client). Ideally, while also checking hashes for known cracks, we update a potfile with every hash and that can be used as a first run against other types of hashes for a potential quick win." - Something to test out, and increase your power bill."

24. Stop Shoddy Academic “Research”

    Papers that are referencing really old projects are called into question. Richard's old project has been dead for years, and the domains point to gambling sites, yet this new academic paper uses it as a reference. We need much better peer review for these papers, such as peers that actually read and check references in the paper, not just the rubber stamping process that it happening today.

25. GitHub – vulncheck-oss/0day.today.archive: An archive of 0day.today exploits

    "0day.today was a long-running public repository of exploits and shellcode. It hosted tens of thousands of PoCs for vulnerabilities affecting a wide range of platforms. In early 2025, 0day.today went offline. Months later it came back but is missing all of its data, effectively erasing over a decade of exploit documentation from the internet. Due to the site's use of anti-bot protection, much of its content was never cached by the Internet Archive, making recovery difficult." - Something to clone as it may come in handy later. You like exploits? I like exploits. We should hang out.

26. Docker makes Hardened Images Catalog affordable for small businesses

    Hardened images should be something that everyone has access to, not just small businesses. Chainguard has some nice Docker images that are hardened, and a free plan as well. Also, creating your own Docker images that are CVE-free is not that difficult.

27. Hacking for Hidden Access And Secret Functions in the Moxa NPort Series

    Pretty neat how jumpering two pins can give you access to the bootloader. Nice hardware hacking tutorial, though nothing earth-shattering, just some backdoor credentials, typical stuff.

28. TecSecurity – tp-link_ax1800

    These are old exploits, not sure how it came up on my feed.

29. The Original Sin of Computing…that no one can fix

    This is such a fun and entertaining video, with the added bonus of being highly educational. You've heard us talk about the bug in the compiler that puts in backdoors, Ken Thompson's work. Youtuber LaurieWired does a great job explaining everything, and even goes on to create her own quine, or self-reproducing code (https://en.wikipedia.org/wiki/Quine_(computing)).

30. ESP32 and Termux

    Not sure why you need to do this from your phone...

Larry Pesce

1. Bodycam Hacking – Luke’s Blog
2. Paged Out!
3. Quick REDUX – LG WebOS TV Path Traversal, Authentication Bypass and Full Device Takeover – SSD Secure Disclosure

   Affected Versions LG WebOS 43UT8050 Vendor Response The vendor has issued an advisory SMR-SEP-2025, available at: https://lgsecurity.lge.com/bulletins/tv in regard to the below described vulnerability Credit The vulnerability was disclosed during our TyphoonPWN 2025 LG Category and won first place. Vulnerability Details The browser-service on WebOS TV opens port 18888 when a USB storage device is

4. Software update bricks some Jeep 4xe hybrids over the weekend
5. Tactility is an operating system for your ESP32, complete with app support
6. From hacker to hero, this startup raised $2.6M to redefine cybersecurity with a conscience – Refresh Miami
7. Satellites Are Leaking the World’s Secrets: Calls, Texts, Military and Corporate Data

Lee Neely

1. Prosecutors seek 7-year prison term for ‘sophisticated’ PowerSchool hacker

   Prosecutors have recommended restitution of over US$14 million and a sentence of seven years in prison for Matthew Lane, the 19-year-old from Worcester, Massachusetts who in May 2025 pleaded guilty to charges of cyber extortion and aggravated identity theft from two unnamed companies, one US telecommunications company and one "cloud-based software company that helped K-12 schools manage student and teacher data," believed to be PowerSchool.

   Lane clearly knew what he was doing, having compromised multiple companies since 2021, and had plans to take his skills to a job at Google. His future now lies along a different path. In the meantime, PowerSchool is in hot water with the state of Texas for falsely claiming "state of the art" security practices, and has acknowledged that the attack would not have been successful if they had employed MFA. All the more resopn to verify third-party security claims. Also, use this to reinforce the argument for using phishing resistant MFA. In short, don't make the attackers job any easier.

2. Cops seize Scattered Lapsus$ Hunters’ BreachForums domain

   The BreachForums domain has once again been taken down. Law enforcement authorities from the US and France have seized the domain on Thursday, October 9, hours before the threat actor group responsible for the theft of data from multiple Salesforce instances had planned to post the stolen information.

   The BreachForums site has been revived more times than 1990s boy bands have made combacks. The efforts this time seem more comprehensive and may truly be the proverbial stake through the heart as they included seizing and destroying the hardware and database backups since 2023. The site had about 340,000 members before the takedown, and facilitated access to sensitive personal information of U.S. citizens.

3. Oracle Patches Another E-Business Suite Vulnerability

   On Saturday, October 11, Oracle released a patch for another vulnerability in its Oracle E-Business Suite. The patch for CVE-2025-61884, a high-severity, remotely exploitable vulnerability, follows just a week after Oracle patched CVE-2025-61882, a critical vulnerability in E-Business Suite that was being actively exploited. According to Oracle's advisory for CVE-2025-61884, the vulnerability affects the Runtime UI of Oracle Configurator, and "may be exploited over a network without the need for a username and password ...[potentially] allow[ing] access to sensitive resources." The advisory says the vulnerability readily affects Oracle E-Business Suite versions 12.2.3 through 12.2.14. Users are urged to apply updates as soon as possible.

4. Reducing notification overload for a quieter browsing experience in Chrome

   In the interest of "a cleaner, more focused browsing experience," Google's Chrome browser is launching a new feature that will allow users to stop receiving notifications from websites they have not visited recently. In a blog post, Chrome Product Manager Archit Agarwal notes that "less than 1% of all notifications receive any interaction from users."

   This is part of Chrome's Safety Check, which was enhanced in September 2024 to  include one-time permissions as well as remove camera/location access on infreqently visited sites. You will be able to review and alter the behavior on the Chrome  - chrome://settings/safetyCheck.

Sam Bowne

1. ‘It’s going to be really bad’: Fears over AI bubble bursting grow in Silicon Valley

   OpenAI is at the centre of a tangled web of deals involving Nvidia, AMD, Microsoft, and Oracle. Some people call these deals "circular financing" or even "vendor financing" - where a company invests in or lends to its own customers so they can continue making purchases. People I've spoken to keep bringing up Nortel - the Canadian telecom equipment-maker that borrowed prolifically to help finance deals for their customers (and thereby artificially boost demand for their wares).

2. AI Data Centers Are an Even Bigger Disaster Than Previously Thought

   AI data centers can't possibly earn enough to justify their cost before they become obsolete.

3. The More Scientists Work With AI, the Less They Trust It

   Scientists expressed less trust in AI than they did in 2024, when it was decidedly less advanced. Concerns include hallucinations, security, and privacy.

4. DDoS Botnet Aisuru Blankets US ISPs in Record DDoS

   The world’s largest and most disruptive botnet is now drawing a majority of its firepower from compromised Internet-of-Things (IoT) devices. It siphons Internet bandwidth from an estimated 300,000 compromised hosts worldwide, mostly consumer-grade routers, security cameras, digital video recorders and other devices operating with insecure and outdated firmware, and/or factory-default settings. By late September, Aisuru was publicly flexing DDoS capabilities topping 22 Tbps. Then on October 6, its operators heaved a whopping 29.6 terabits of junk data packets each second at a targeted host.

5. AI models can acquire backdoors from surprisingly few malicious documents

   250 malicious documents can successfully poison a model with 13 billion parameters, trained on 260 billion tokens. This raises a serious concern--how can we prevent this attack? AI requires a massive training dataset, but how can it possibly be that clean?

6. Introduction to Data Poisoning: A 2025 Perspective

   A Nature Medicine study found that replacing just 0.001% of training tokens in a medical dataset with misinformation caused models to generate 7–11% more harmful completions. This reinforces the previous article's results--AI training data must be very clean to avoid poisoning, and it is not obvious at all how to find data that clean.

7. Hobble your AI agents to prevent them from hurting you too badly

   Everyone is rushing to deploy AI agents, which can take actions for the user, such as sending emails or making online purchases. But the AIs cannot be trusted. The best risk mitigation is to limit what the agents can do: Don't give them access to file deletion commands. Don't let them open arbitrary network ports.

8. Insecure Mobile VPNs: The Hidden Danger

   They request dangerous permissions, driven by a desire to collect data for monetization. They also exported activities or content providers without proper safeguards, allowing other apps on the device to launch them or query their data. Similarly, system-level calls such as Runtime.exec() can be abused to execute arbitrary commands or bypass platform security features.

9. Hackers can steal 2FA codes and private messages from Android phones

   A malicious app without system privileges can read data off the phone screen, by drawing over that content and measuring the timing of the GPU's graphical data compression. The attack may be too slow to collect a useful 2FA token within its 30 second lifetime. Google has issued a patch which partially mitigates this behavior.